Relation between voltage, temperature, clock speed and impact on chip longevity

[JamesWatt1]

One thing I have been curious about is how voltage, temperature, and clock interact as well as the impact of each on chip longevity. This my understanding based on some googling.

Is this accurate? I would also appreciate it if you would fill in the rates where I leave them ambiguous (e.g., I say increases and it actually increases exponentially).

What is the relationship between voltage, temperature, and clock speed?
As voltage increases, stable clock speed increases
As voltage increases, temperature increases exponentially
As temperature increases, voltage decreases.
As temperature increases, stability at a given clock speed decreases.
As clock speed increases, voltage is constant and temperature increases linearly.

What is the impact of voltage, temperature, and clock speed on chip life?
As voltage increases, chip life decreases exponentially
As temperature increases, chip life decreases exponentially
As clock increases, chip life decreases proportionally

I've read that electro-migration, hot carrier degredation, and oxide breakdown are the main causes of chip degredation (http://www.rockcomputerma.com/the-cause-of-life-limiting-cpu-cpu-life-clock-speed.html). Do you have any idea the rough percentages of chip deaths due to each?

 

[TecHNooB]

As temperature increases, voltage decreases.

Can you elaborate on this? As far as I know, heating a resistor causes its resistance to increase while heating a semiconductor causes its resistance to decrease. But a good power supply is supposed to hold the voltage.

I think there's an empirical formula which goes something like Power ~ Constant * Voltage * Frequency^2

If I have that formula correct, it looks like it would suggest that temp goes up linearly with voltage.

 

[fail]

As temperature increases, voltage decreases.

Can you elaborate on this? As far as I know, heating a resistor causes its resistance to increase while heating a semiconductor causes its resistance to decrease. But a good power supply is supposed to hold the voltage.

I think there's an empirical formula which goes something like Power ~ Constant * Voltage * Frequency^2

If I have that formula correct, it looks like it would suggest that temp goes up linearly with voltage.

fail

 

[fail]

..

 

[TecHNooB]

fail

hehe, i just looked up the formula. P = C*V^2*F. Makes a bit more sense now

 

[klinc]

Microprocessors heat due to Joule effect, which is the process of transforming electrical energy into heat. Inside the CPU there are several wires (conductors) in charge of its internal interconnections. The Joule effect appears due to the shock between electrons and the conductor ion mesh, leading to an increase in the temperature of the conductor.
The heat generated by an electronic device needs to be removed as soon as possible; otherwise its internal temperature will increase. If the device gets too hot internally, its internal circuits can be damaged, thing that we don’t want, of course.
The maximum CPU temperature is usually written on its body in a coded format – i.e., a letter added somewhere indicates what the CPU maximum temperature is. This code isn’t standardized; it varies according to the CPU. On the CPU datasheet, which is available at the manufacturer’s website, there is a section that explains the coding used on the CPU, which includes the CPU maximum temperature.
This temperature is the maximum temperature the CPU can work without burning. The lower the CPU temperature, the better. Good quality CPU coolers and the correct use of thermal grease will make your CPU to work way below its maximum rated temperature.

Posting someone else's work without citation is plagiarism. Posting this much of someone else's work is copyright infringement. Please do not do it again.
Admin allisolm

 

[Paperdoc]

"As voltage increases, stable clock speed increases." The correlation exists, but the causation is not clear here. The reason has to do with circuit response time inside the CPU (or other chip). The signals in digital circuits are constantly being switched from on to off and back. The change from one state to another is never instantaneous - it takes time, although short. So after each change is initiated, you have to wait for a couple of "response times" for the circuit output voltage to rise (or fall) to the level necessary to represent accurately the "on" (or "off") state. If you try to speed up the circuit clock and make all the state changes happen in shorter times, the wait period may not be enough for that stable state to be reached before it is "read" and then changed again, and this produces errors. One way to beat that is to raise the supply voltage to the circuit so that the required voltage is reached in a shorter time (because ultimately the voltage will actually end up higher). Then you can "get away with" running the clock faster and waiting for shorter times after each stat change.

"As voltage increases, temperature increases exponentially." Not really exponentially, but sort of. Simply from the standpoint of the generation of heat by the circuit, increasing the voltage by a certain fractional factor, say, x (like, 1.10, a 10% increase) will also increase the current that flows through the circuit by that same factor. This means that the heat generated (power dissipated) in that circuit, which is the product of voltage time current, increases by the square of x (in the example here, 1.1 squared, or 1.21 - thus 21% higher heat generation for a 10% voltage increase. But the effect gets magnified a bit because the rate of heat flow out of the circuit is limited by the thermal conductivity of the materials, so the temperature right at the circuit active elements on the chip rises even more than you might guess from that power increase. The results is not exactly an exponential relationship between voltage and temperature, but it is very significant.

"As temperature increases, voltage decreases." Well, it's more like: if your temperatures get too high, you had better reduce the voltage supplied to the circuit to prevent disastrously high temperatures from destroying the chip.

"As temperature increases, stability at a given clock speed decreases." Not directly. BUT if your temperature increase is a result of having raised the clock speed without raising the voltage, you may well get to unstable operation. See comments on the first item and response times.

"As clock speed increases, voltage is constant and temperature increases linearly." If you increase clock speed without increasing the chip's supply voltage, the chip temperature will increase. That is because the heat per cycle being generated does not change, but the heat per second does - more cycles per second. And the thermal conductivity of the materials is fixed, so the material heats up as more heat per second is generated. May not be exactly linear, but the effect is less severe than the effect of voltage increases. However, as above, people who increase clock speed often follow that with an increase in supply voltage to regain stability, and the combined effect is significant. That's why overclockers spend so much effort on heat removal.

 

[pm]

I've read that electro-migration, hot carrier degredation, and oxide breakdown are the main causes of chip degredation (http://www.rockcomputerma.com/the-ca...ock-speed.html). Do you have any idea the rough percentages of chip deaths due to each?

There's a couple of new ones: PMOS BTI (PMOS Bias Temperature Instability) and TDDB (time dependent dielectric breakdown, although this could be a subset of "oxide breakdown").

As far as percentages, I used to do FA (Failure Analysis), and I tended to see everything fall into about a dozen buckets. At the time that I did this job, the biggest failures were to electromigration but usually in areas that were problematic due to lithography issues caused by not-strict-enough design rules. We'd fix those, and then the next bunch tended to be PMOS BTI and TDDB. PMOS BTI can be weeded out and even fixed in burn-in with a good burn-in flow, so then that left oxide breakdown and hot-e. But both of these were hard ones to do FA on so sometimes I wasn't sure if it was more of a guess. So, back to percentages, it was all over and moved through time as the design and fab became more mature.